Neural Network-based Prediction Modeling for External Labeling

External labeling algorithms have been investigated and applied in several fields over the last twenty years. This paper develops a data-driven prediction modeling method for external labeling, which exploits the adoption of deep learning in external labeling. In this method, the multilayer perceptron (MLP) network is employed to identify the hidden pattern between the model components and label information. The padding technique is modified and used in the input layer of the MLP network to make the network adaptive for objects with a different number of components to be annotated. The MLP network is evaluated using virtual models with different numbers of components and contours. Preliminary experimental results indicate that the MLP network can achieve real-time external labeling under changing viewpoints, meet the requirements for external labeling, adapt to various contours, including convex hulls and concave hulls, and avoid label jumping. Moreover, the effectiveness of the proposed method is also demonstrated by performance tests on various virtual models with labeling accuracy of over 90%.